Quantitative Identification Model Of Polyester-Cotton Fabric And Its Visualization Based On Hyperspectral And Imaging Technique

There are many methods for quantitative detection of textile materials,but each method has its own defects,such as damage samples,low identification efficiency,excessive dependence on operator experience,and inability to achieve real-time large-scale detection.Therefore,in order to respond to the national call for conservation and green environmental protection,this topic is mainly based on hyperspectral imaging acquisition system,from the near-infrared band hyperspectral reflectance curve data,and hyperspectral image development:a study on textile materials The hyperspectral image acquisition conditions were used to characterize the spectral reflectance curves of different components of polyester-cotton fabrics.The hyperspectral image data preprocessing software was developed.Based on the software,different pretreatment methods were analyzed.Based on the influence of the spectral curve,a quantitative identification model of polyester-cotton fabric based on near-infrared hyperspectral data was constructed and optimized.Based on the edge operator detection method,an algorithm for automatically extracting the region of interest of hyperspectral image was studied.The image classification method displays the classified and identified images in the form of a false color composite map,thereby realizing visualization of quantitative identification of hyperspectral images.The specific research contents and conclusions are as follows:(1)In order to establish the conditions for the acquisition of hyperspectral imagery for textile materials,the subject compared the effects of different illumination conditions,focusing conditions,movement speed of the stage and the number of folded layers on the hyperspectral image.The research shows that the hyperspectral imaging acquisition system can eliminate the influence of external illumination factors through black-and-white calibration;poor lens focusing state will result in blurred hyperspectral image and unclear edge of the sample,which is not conducive to the visualization of subsequent image classification,but for pixels.The point spectral data has no effect;the movement speed of the stage too fast or too slow will cause the sample in the hyperspectral image to be elongated or laterally elongated,but has no effect on the average spectral data in the effective area of the sample;For larger fabrics,the spectral data in the effective area of the sample in a layer of fabric is affected by the background of the stage,and there is no effect on the fabric state of more than two layers(2)In order to verify the near-infrared band involved in the use of hyperspectral images,different amounts of polyester-cotton fabrics can be quantitatively analyzed.On the basis of the established hyperspectral image acquisition conditions,polyester,cotton and different content of polyester cotton are characterized.It can be seen from the study that there are three characteristic peaks in the band of different content of polyester-cotton fabrics,which exist at 1129,1160 and 1490 nm respectively.Due to the characteristic peak of polyester at 1129 nm,the peak intensity of the characteristic peak gradually decreases with the decrease of the proportion of polyester content in the two-component fabric;the cotton has a strong characteristic peak at 1490 nm,with the two-component fabric An increase in the proportion of cotton content leads to a gradual increase in the peak intensity of the characteristic(3)Based on the principle of hyperspectral image data storage,the hyperspectral image data preprocessing software was developed independently.The software can realize the functions of hyperspectral image display,spectral curve generation and spectral data preprocessing.It can improve the efficiency of establishing a quantitative model.(4)In order to establish the partial least squares regression model with the best identification efficiency,the self-developed hyperspectral image data preprocessing software was used to compare the different preprocessing methods and the training set sample selection method.The least squares method is used to quantify the regression model,and the model is optimized by the complex correlation coefficient and the predicted root mean square.The research shows that the first derivative processing is the best pretreatment method,and the SPXY sample set selection method is the most suitable selection method for this study.When the number of principal components is 8,the model is the best.(5)In order to realize the quantitative detection of polyester-cotton fabrics capable of high-volume,on-line sorting,a method for visualizing quantitative identification of polyester-cotton fabrics was studied based on hyperspectral images.It is found that due to the different warp and weft yarns of polyester-cotton fabrics,the image classification for pixel points cannot achieve the purpose of quantitative identification.Therefore,based on the edge detection operator,median filtering,image binarization and threshold method,an algorithm for automatically extracting the effective area of the sample in the image is studied.Under the premise of establishing standard polyester and cotton fabric color cards,the PLS quantitative identification model is used to predict the unknown samples to realize the identification of the content of polyester-cotton fabric components.